Re: 10g RAC design options

From: DA Morgan <damorgan_at_psoug.org>
Date: Thu, 07 Jul 2005 06:41:28 -0700
Message-ID: <1120743717.75517@yasure>

Rodrick Brown wrote:

>>I am confused by what you have written. Are you, perhaps, confusing RAC
>>with DataGuard? DataGuard has active-active but with RAC the nodes are
>>either active or they don't exist.

>
> Humm not sure how you missed the point that was an example of the current
> environment and limitations i'm not confusing dataguard and RAC read again
> carefully.

Perhaps by reading the Subject "10g RAC". ;-)

>>Also with RAC it would be a horrible waste of money to build with  CPU
>>machines. The best possible value comes from 2 CPU nodes.

>
> My databases are pretty big and CPU intensive, has this been documented some
> where?

Not to be insulting but read the RAC Concepts books. The number of CPUs in a node is essentially irrelevant to how CPU intensive your application is. There is a difference, but when economics are added in not a significant one, between 16 CPUs put together as as 8x2 vs 4x4. But there is a huge difference in cost.

>>Unless you have some reason to want to spend more money and get less
>>performance you are, apparently without serious research, keying in on
>>some of the slowest and most expensive options available. Why?

>
> Scaling horizontally is much easier and cheaper than scaling veritcally
> thats the main reason why I want to do away with Large database servers and
> move to RAC so i'm not seeing your logic here.

Actually I saw yours and what I wrote was apparently not clear. I am in full agreement with respect to going to RAC for the reasons you state and more. Going to RAC with 4CPU machines. Going to RAC with Sun. Or going to RAC with RHEL is quite another matter.

You are making decisions without doing the research. There are far better options. And by better I mean faster, less expensive, and with fully supported OS.

>>>Most of the staff is very Solaris savvy which is why Solaris looks like a 
>>>much better fit, I do not want to go with a sparc based solution because 
>>>of the high cost of Oracle licenses and slow CPU performance compared to 
>>>x86.
>>
>>Please explain what you mean by "Solaris savvy"? If they know UNIX they
>>know UNIX. And yes there are minor differences between Solaris and HP/UX
>>and AIX and Linux and FreeBSD but they are so insignificant that any
>>competent (and I emphasize competent) staff should be able to pick up
>>any other UNIX flavour with a week's training.

>
> Wrong. Linux is nothing like Solaris, yes tons of features overlap but when
> it comes to finding and fixing kernel bugs and bottle necks you can easily
> see the difference between UNIX flavors.

Nothing like? Are you serious? The differences are the differences between purchasing a Ford or a Toyota. How hard is it for a mechanic that knows one to learn how to repair the other?

>>Choosing Solaris because it is what you know is like sticking with COBOL
>>because it is what you know. I'd rethink this concept.

>
> General human nature people tend to use with what they are most familiar
> with.

Can't disagree. But that isn't a good justification for spending 200-300% more money to get hardware with poor performance.

>>>Another concern; uptime have been almost 99.999% over the past 3+ years 
>>>my current solution has been deployed, how are most people deploying RAC? 
>>>Do you cluster between RAC using VCS? I've read that their can be certain 
>>>situations where patching RAC would require every node being offline, in 
>>>this environment this is not a suitable option.
>>
>>Absolutely not. Assuming 10g we cluster RAC using ASM managed RAW. I
>>can't imagine why anyone would pay Veritas for something provided by
>>Oracle for free nor why anyone would choose to deal with two
>>finger-pointing vendors when they could deal with only one.

>
> See the previous quote, I can honestly say Veritas products are very stable
> and makes life a hell of alot easier when dealing with lots of storage and
> want a truly HA setup, I have no experience with ASM but about 5+ years
> using Vertias tools, I dont have anything against learning something new,
> which is why I'm posting here to see what others are doing.

I can honestly say I agree with you but the product is totally obsolete. ASM does things that Veritas can only dream about ... things that are of huge value to an Oracle shop.

Surely you didn't go into IT thinking you could not learn new things and keep using the same products year-after-year and not become as useful as an RPG or ALGOL specialist. ;-)

If you don't want one of my students coming up behind you and taking your job by offering better skills for less money I'd suggest embracing change. And that means doing an honest evaluation of all UNIX based operating systems without prejudice about current skill sets and doing an honest evaluation of ASM.

You shouldn't be handicapping yourself, or your employer, by digging in your heels and trying to stay with what you currently know. Sun is good. I have a Sun machine on the desk just 18 inches from me right now. But I wouldn't recommend one in an Oracle shop today, and likely won't again tomorrow, until Sun gets a reality check from the marketplace.

-- 
Daniel A. Morgan
http://www.psoug.org
damorgan_at_x.washington.edu
(replace x with u to respond)

Received on Thu Jul 07 2005 - 08:41:28 CDT